LETTER

doi:10.1038/nature13586

Quantum imaging with undetected photons

Gabriela Barreto Lemos1,2, Victoria Borish1,3, Garrett D. Cole2,3, Sven Ramelow1,3{, Radek Lapkiewicz1,3 & Anton Zeilinger1,2,3

Information is central to quantum mechanics. In particular, quantum

interference occurs only if there exists no information to distinguish
between the superposed states. The mere possibility of obtaining
information that could distinguish between overlapping states inhibits quantum interference1,2. Here we introduce and experimentally
demonstrate a quantum imaging concept based on induced coherence without induced emission3,4. Our experiment uses two separate
down-conversion nonlinear crystals (numbered NL1 and NL2), each
illuminated by the same pump laser, creating one pair of photons
(denoted idler and signal). If the photon pair is created in NL1, one
photon (the idler) passes through the object to be imaged and is overlapped with the idler amplitude created in NL2, its source thus being
undefined. Interference of the signal amplitudes coming from the
two crystals then reveals the image of the object. The photons that
pass through the imaged object (idler photons from NL1) are never
detected, while we obtain images exclusively with the signal photons
(from NL1 and NL2), which do not interact with the object. Our experiment is fundamentally different from previous quantum imaging techniques, such as interaction-free imaging5 or ghost imaging69,
because now the photons used to illuminate the object do not have
to be detected at all and no coincidence detection is necessary. This
enables the probe wavelength to be chosen in a range for which suitable detectors are not available. To illustrate this, we show images of
objects that are either opaque or invisible to the detected photons.
Our experiment is a prototype in quantum informationknowledge
can be extracted by, and about, a photon that is never detected.
The conceptual arrangement of our imaging technique, based on a
quantum interference experiment3,4 by Zou, Wang and Mandel, is illustrated in Fig. 1. A pump beam (green) divided by a 50:50 beam splitter
(BS1) coherently illuminates two identical nonlinear crystals, NL1 and
NL2, where pairs of collinear photons called signal (yellow) and idler
(red) can be created (jcijdi in NL1 and jeijf i in NL2). The idler amplitude created in NL1 reflects at the dichroic mirror D1 into spatial mode
d, and signal amplitude passes into spatial mode c. The idler passes
through
  coefficient T and phase shift
ptransmittance
  the object
 O of real
c: cis dii ?Teic cis dii z 1{T 2 cis wii, where for simplicity we lump
all lost idler amplitude into a single state wii (here subscripts s and i
represent signal and idler). By reflection at dichroic mirror D2, the
idler
from

 NL1 aligns perfectly with idler amplitude produced at NL2,
di ?f i . The state at the grey dotted line is thus
i
i
p  i

 
1 h
p Teic cis zeis f ii z 1{T 2 cis wii
1
2
The idler is now reflected
at the dichroic mirror D3 and discarded. The

signal states cis and eis are combined at the 50:50 beam splitter BS2. The


detection probabilities at the outputs gis and his , obtained by ignoring
(tracing out) the idler modes, are
1
Pg=h ~ 1+T cos c
2
2
Thus, fringes with visibility T can be seen at either output, even though the
signals combined at BS2 have different sources4,10. These fringes appear in

the signal single photon counts; the idlers are not detected. No coincidence
detection is required.
The peculiar feature of this interferometer is that no detected photon
has taken path d. Yet, in our experiment, it is precisely here where we put
the object to be imaged. The key to this experiment is how the signalsource information carried by the undetected idler photon depends on
T. For,
 if T5 0, an idler detected after D3, coincident with a signal count
at gis or his , would imply the signal source was NL2. Detection of a
signal photon without a coincident idler would imply the signal source
was NL1. This which-source information destroys interference because
it makes the quantum states overlapping at BS2 distinguishable. If
T 5 1, the idler photon carries no which-source information. The signal states overlapped at each output of BS2 are then indistinguishable;
thus the interference term in equation (2) appears. The above arguments
are valid even though the idler photons are not detected, for it is only
the possibility of obtaining which-source information that matters in
this experiment.
Our experiment has a connection to interaction-free measurements11,12.
Note that Ph ~0 if no object is placed in the set-up (T 5 1 and c 5 0).
Now insert an opaque object (T 5 0) so thatPh . 0, and monitor the
idler reflection from D3. Coincident counts in his and the idler detector
reveal that the object is present even though no photon interacted with
the object. With our set-up it is thus possible to realize non-degenerate
interaction-free imaging.
With O and D2 removed,
  equation
  (1) would be an ordinary two
particle entanglement13, cis dii zeis  f ii . With them in, dii ?Teic  f ii ,
which creates equation (1). A normal two-particle entanglement has
changed into an interesting single-particle superposition, which is especially rich when T and c are transverse-position dependent.
We expand the conceptual arrangement of Fig. 1 into an imaging
system (Fig. 2). We replace the photon counters with cameras sensitive
to single photons and the uniform object with one bearing features, that
is, T 5 T(x, y) and c 5 c(x, y) depend on transverse position (x, y). Our
source produces spatially entangled photon pairs14,15. Sharp spatial correlations between signal and idler in the object plane and confocal lens
systems16 (see Methods) guarantee a point-by-point correspondence between the object plane and the detector surface on the camera.
The intensity image (non-constant transmittance) is due to transverseposition-dependent which-source information carried by the undetected
idler photons. The phase image is of a different nature: it is due to the fact
that the position-dependent phase shift
 on the
 idler
  inpathic d is
 photons
17 
ic 
 

actually
 passed to the signal; that is , cis Te f ii z eis f ii ~ Te cis
zeis  f ii . Remarkably, the idler beam  f ii alone does not even carry the
phase pattern, and without detection in coincidence it could not be used
to obtain the phase image18,19.
We will now show images obtained by detecting 810-nm photons
with a camera capable of single-photon sensitivity at this wavelength,
when three different objects are illuminated by 1,550-nm photons, to
which our camera is blind (see Methods). First, a cardboard cut-out
placed into the path D1D2 is imaged. Next, we show that a positiondependent phase shift produces an image even when the object is opaque
(an etched silicon plate) or invisible (etched silica plate) at the detection

Institute for Quantum Optics and Quantum Information, Austrian Academy of Sciences, Boltzmanngasse 3, Vienna A-1090, Austria. 2Vienna Center for Quantum Science and Technology (VCQ), Faculty of
Physics, University of Vienna, A-1090 Vienna, Austria. 3Quantum Optics, Quantum Nanophysics, Quantum Information, University of Vienna, Boltzmanngasse 5, Vienna A-1090, Austria. {Present address:
Cornell University, 159 Clark Hall, 142 Science Drive, Ithaca, New York 14853, USA.
2 8 AU G U S T 2 0 1 4 | VO L 5 1 2 | N AT U R E | 4 0 9

2014 Macmillan Publishers Limited. All rights reserved

RESEARCH LETTER
BS1

D1

NL1

300

800

c
Pixels

600

200

400
100
200

O
0

100

200

300
Pixels

D3
BS2

f
e
NL2

D1

NL1

200

800

100

400

0
0

100 200
Pixels

300

900

200

600

100

300

300

100 200
Pixels

thus it is impossible to realize transmission imaging by illuminating the

silicon with 810-nm photons. However, silicon is highly transparent at
1,550 nm and when we place the object in path D1D2, the difference in
optical path length for the etched and non-etched regions corresponds to
a relative phase shift of p. Even though our camera is blind to 1,550-nm
light, the image is seen by detecting 810-nm photons at the output of BS2
(Fig. 4a).
Finally, Fig. 5a shows the image of a fused silica (SiO2) plate etched
with a pattern that is invisible at the detection wavelength (details are
given in the Methods section). We take advantage of the flexibility of our
source to obtain collinear non-degenerate down-conversion at 820 nm
a

2,500
200

Pixels

2,000
1,500
100

1,000

L4

500

532 nm

0
L3
1

M1

L3

100

200

300

500

Pixels

400
200
0

BS
D5

L2 D2

400

z (nm)

2
L2

L4

Figure 3 | Intensity imaging. a, Inside the cat, constructive and destructive

interference are observed at the outputs of BS when we placed the cardboard
cut-out shown in b in the path D1D2. Outside the cat, idler photons from
NL1 are blocked and therefore the signals do not interfere. c, The sum of the
outputs gives the intensity profile of the signal beams. d, The subtraction of
the outputs leads to an enhancement of the interference contrast, as it highlights
the difference between constructive and destructive interference.

D4

HWP

300

Counts

PBS

1,200

Counts

wavelength. The images obtained with an electron multiplying charge

coupled device (EMCCD) camera show single (non-heralded) counts
per pixel (16 mm 3 16 mm) obtained in an exposure time of 0.5 s with an
electron multiplying gain factor of 20. The visibility achieved in the
experiment is 77% (see Methods for details).
Figure 3a shows the beamsplitter output when a cardboard cut-out
(illustrated in Fig. 3b) is inserted in the path D1D2. Constructive interference is seen at one output of the beam splitter and destructive interference is observed in the other output. Interference only occurs in the
region corresponding to the idler beam transmitted through the shape
cut out of the cardboard, as seen in the sum and difference of the complementary images, shown in Fig. 3c and d, respectively. The sum of the
two outputs of the beamsplitter gives the featureless intensity profile of
the signal beams, demonstrating that the signal beams, while carrying the
intensity information, are not absorbed at all by the mask.
In Fig. 4a, we show the image of an etched 500-mm-thick silicon plate;
the plate is shown in Fig. 4b (see Methods section for details of the silicon
plate and the etching process). Silicon is opaque to illumination at 810 nm,

600

d
300

Counts

Figure 1 | Schematic of the experiment. Laser light (green) splits at beam

splitter BS1 into modes a and b. Beam a pumps nonlinear crystal NL1, where
collinear down-conversion may produce a pair of photons of different
wavelengths called signal (yellow) and idler (red). After passing through the
object O, the idler reflects at dichroic mirror D2 to align with the idler produced
in NL2, such that the final emerging idler | fi does not contain any information
about which crystal produced the photon pair. Therefore, signals | cs and
| es combined at beam splitter BS2 interfere. Consequently, signal beams | gs
and | hs reveal idler transmission properties of object O.

WPs

500

Pixels

D2

400

Pixels

L1

3 mm

Counts

NL2
L5

L6

4
3

Filters

3
2

1
1

mm

Figure 2 | Experimental set-up. A continuous-wave 532-nm laser (green)

illuminates crystals NL1 and NL2. Wave plates (WPs) adjust the relative phase
and intensity of the outputs of the polarizing beam splitter (PBS). The dichroic
mirror D1 separates down-converted 810-nm (yellow) and 1,550-nm (red)
photons. The 1,550-nm photons are transmitted through the object O and sent
through NL2 by dichroic mirror D2. Lenses image plane 1 onto plane 3, and
plane 2 onto the EMCCD camera. A 50:50 beam splitter (BS) combines the
810-nm beams. Dichroic mirrors D1, D2, D4 and D5 transmit the pump.

y(

EMCCD

0
0

m)

x (m

Figure 4 | Phase image of an object opaque to 810-nm light. a, Detection of

810-nm photons at both outputs of BS when a silicon plate (opaque to 810-nm
light) with a 3-mm-tall etched cat (b) was introduced in path D1D2. b, Threedimensional rendering of the etch design overlaid with stylus profilometer
scans (blue points) of the actual etch depth.

4 1 0 | N AT U R E | VO L 5 1 2 | 2 8 AU G U S T 2 0 1 4

2014 Macmillan Publishers Limited. All rights reserved

LETTER RESEARCH
a

400
400
300

200

200

100

100
0

400

400

300

300

200

200

100

100

0
0

100

200

300

400

500

600

700

800

Counts

Pixels

Counts

Pixels

300

Pixels

b
2
z (m)

1
3
y

2
(m

1
0

m
x (m

that information can be obtained about an object without detecting the

photons that interacted with the object. Knowing the two-photon state,
one can obtain information about an object. It has not escaped our attention that, on the other hand, by knowing the object, one could obtain
information about the quantum state without detecting it.

METHODS SUMMARY
A detailed schematic of our imaging set-up is shown in Fig. 2. A 532-nm linearly
polarized Gaussian pump laser beam focused by lens L1 on plane 1 is divided at a
polarizing beam splitter (PBS) and coherently illuminates two identical periodically
poled potassium titanyl phosphate (ppKTP) crystals, NL1 and NL2. The PBS plus
wave plates (WPs) are used to control the relative amplitudes and phases between the
reflected and transmitted pump beams. With an extra half-wave plate (HWP) in the
reflected beam, both beams have the same polarization. The 1,550-nm idler amplitude produced at NL1 is reflected by dichroic mirror D1, through which the 810-nm
signal and the pump are transmitted. Dichroic mirror D4 transmits 532-nm light
and reflects 810-nm light. A long-pass filter (not shown in the figure) placed directly
before the object O blocks any residual 532-nm or 810-nm light. The 1,550-nm
amplitude from NL1 illuminates the object O and is then overlapped with the pump
beam at dichroic mirror D2 that transmits 532-nm light and reflects 1,550-nm light.
Lens pairs L2L29, L3L39, and L4L49 image plane 1 onto plane 3, thereby ensuring that pump, idler and signal, respectively, are identical in these planes, thus contributing to obtain high interference visibility21 (see Methods). Lenses L5 and L6
together with L39 and L49 image object plane 2 onto the camera surface.
The 810 6 1.5 nm photons are detected (without heralding) in both outputs of the
BS using an EMCCD camera that exhibits single-photon sensitivity at 810 nm, but
has a negligible response at 1,550 nm.
Online Content Methods, along with any additional Extended Data display items
and Source Data, are available in the online version of the paper; references unique
to these sections appear only in the online paper.
Received 26 January; accepted 11 June 2014.
1.

Figure 5 | Phase imaging of a 2p step at 820 nm. a, The top picture was taken
with the object (shown in b) placed in the 820-nm beam between L4 and L49;
in the bottom picture, the object was placed in the 1,515-nm beam in path
D1D2. b, Three-dimensional rendering of the design overlaid with stylus
profilometer scans (blue dots) of the actual etch depth.

2.
3.
4.
5.

and 1,515 nm (see Methods). The object (Fig. 5b) has an etch depth of
1,803 nm, imparting a relative phase shift of ,2p for 820-nm light. Thus
the object is invisible when placed in the path of the detected photons
between L4 and L49 (top of Fig. 5a). This same etch depth gives an ,p
phase step for 1,515-nm light, so when this same object is placed in the
path D1D2 of undetected photons, an image seen in the contrast of
constructive to destructive interference is retrieved in the 820-nm outputs (bottom of Fig. 5a).
In summary, we have presented a quantum system for intensity and
phase imaging where the photons that illuminate the object are not
detected and the photons that are detected do not illuminate the object.
We image objects that are either opaque or invisible at the detection
wavelength (near-infrared) by illuminating three different objects with a
wavelength to which our detector is blind. This experiment is fundamentally different to ghost imaging69 as it relies on single-photon interference and does not require coincidence detection. Furthermore, our
technique could be used for non-degenerate interaction-free imaging,
with potential applications spanning biological imaging to the inspection of integrated circuits. Our system can realize grey-scale intensity
or phase imaging, and it can be modified in order to measure spectral
features (spectral imaging)20.
We have demonstrated that our technique does not require the laser or
the detector to function at the same wavelength as that of the light probing
the object. Additionally, any nonlinear process can be used as a source, and
this provides flexibility in the wavelength range for both detection and
illumination of the object. In particular, in spontaneous parametric downconversion (as used here), the only absolute restriction is that the sum of
the two photon energies equals that of the pump photons. We have shown

6.

7.
8.
9.
10.
11.
12.
13.
14.

15.
16.

17.

18.

19.

20.

Feynman, R. P., Leighton, R. B. & Sands, M. The Feynman Lectures on Physics Vol. III,
Chs 1 and 3 (Addison-Wesley, 1964).
Mandel, L. Coherence and indistinguishability. Opt. Lett. 16, 18821883 (1991).
Zou, X. Y., Wang, L. J. & Mandel, L. Induced coherence and indistinguishability in
optical interference. Phys. Rev. Lett. 67, 318321 (1991).
Wang, L. J., Zou, X. Y. & Mandel, L. Induced coherence without induced emission.
Phys. Rev. A 44, 46144622 (1991).
White, A. G., Mitchell, J. R., Nairz, O. & Kwiat, P. G. Interaction-free imaging. Phys.
Rev. A 58, 605613 (1998).
Abouraddy, A. F., Stone, P. R., Sergienko, A. V., Saleh, B. E. A. & Teich, M. C.
Entangled-photon imaging of a pure phase object. Phys. Rev. Lett. 93, 213903
(2004).
Gatti, A., Brambilla, E. & Lugiato, L. Quantum imaging. Prog. Opt. 51, 251348
(2008).
Pittman, T. B. et al. Two-photon geometric optics. Phys. Rev. A 53, 28042815
(1996).
Aspden, R. S., Tasca, D. S., Boyd, R. W. & Padgett, M. J. EPR-based ghost imaging
using a single-photon-sensitive camera. New J. Phys. 15, 073032 (2013).
Wiseman, H. M. & Mlmer, K. Induced coherence with and without induced
emission. Phys. Lett. A 270, 245248 (2000).
Elitzur, A. C. & Vaidman, L. Quantum mechanical interaction-free measurements.
Found. Phys. 23, 987997 (1993).
Kwiat, P., Weinfurter, H., Herzog, T., Zeilinger, A. & Kasevich, M. A. Interaction-free
measurement. Phys. Rev. Lett. 74, 47634766 (1995).
Horne, M. in Experimental Metaphysics Vol. 1. (eds Cohen, R. S., Horne, M. & Stachel,
J.) 109119 (Kluwer Academic, 1997).
Howell, J. C., Bennink, R. S., Bentley, S. J. & Boyd, R. W. Realization of the EinsteinPodolsky-Rosen paradox using momentum-and position-entangled photons
from spontaneous parametric down conversion. Phys. Rev. Lett. 92, 210403
(2004).
Walborn, S. P., Monken, C. H., Padua, S. & Souto Ribeiro, P. H. Spatial correlations in
parametric down-conversion. Phys. Rep. 495, 87139 (2010).
Tasca, D. S., Walborn, S. P., Souto Ribeiro, P. H., Toscano, F. & Pellat-Finet, P.
Propagation of transverse intensity correlations of a two-photon state. Phys. Rev. A
79, 033801 (2009).
Horne, M. A., Shimony, A. & Zeilinger, A. Two particle interferometry. Phys. Rev. Lett.
62, 22092212 (1989); Two particle interferometry. Nature 347, 429430
(1990).
Ribeiro, P. H. S., Padua, S., Machado da Silva, J. C. & Barbosa, G. A. Controlling the
degree of visibility of Youngs fringes with photon coincidence measurements.
Phys. Rev. A 49, 41764179 (1994).
Abouraddy, A. F., Stone, P. R., Sergienko, A. V., Saleh, B. E. A. & Teich, M. C.
Entangled-photon imaging of a pure phase object. Phys. Rev. Lett. 93, 213903
(2004).
Zou, X. Y., Grayson, T. P. & Mandel, L. Observation of quantum interference effects
in the frequency domain. Phys. Rev. Lett. 69, 30413044 (1992).
2 8 AU G U S T 2 0 1 4 | VO L 5 1 2 | N AT U R E | 4 1 1

2014 Macmillan Publishers Limited. All rights reserved

RESEARCH LETTER
21. Grayson, T. P. & Barbosa, G. A. Spatial properties of spontaneous parametric
down-conversion and their effect on induced coherence without induced emission.
Phys. Rev. A 49, 29482961 (1994).
Acknowledgements We thank M. Horne for reading the manuscript, clarifying
suggestions and many discussions, P. Enigl for designing the figures for the objects,
D. Greenberger and S. von Egan-Krieger for discussions, and C. Schaeff for equipment
loans. Microfabrication was carried out at the Center for Micro- and Nanostructures
(ZMNS) of the Vienna University of Technology. We acknowledge D. Ristanic for
assistance with cryogenic Si etching and M. Schinnerl for contact mask production.
AW) through a fellowship
G.B.L. was funded by the Austrian Academy of Sciences (O
from the Vienna Center for Science and Technology (VCQ). S.R. is funded by an EU
Marie Curie Fellowship (PIOF-GA-2012-329851). This project was supported by OAW,
the European Research Council (ERC Advanced grant no. 227844 QIT4QAD, and SIQS

grant no. 600645 EU-FP7-ICT), and the Austrian Science Fund (FWF) with SFB F40
(FOQUS) and W1210-2 (CoQus).
Author Contributions A.Z. initiated this research. G.B.L., V.B., R.L., S.R. and A.Z.
designed the experiment. G.B.L., V.B. and R.L. carried out the experiment. G.D.C.
fabricated the silicon and silica phase masks. All authors contributed to the writing of
the manuscript.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Readers are welcome to comment on the online version of the paper.
Correspondence and requests for materials should be addressed to
G.B.L. (gabriela.barreto.lemos@univie.ac.at) and A.Z.
(anton.zeilinger@univie.ac.at).

4 1 2 | N AT U R E | VO L 5 1 2 | 2 8 AU G U S T 2 0 1 4

2014 Macmillan Publishers Limited. All rights reserved

LETTER RESEARCH
METHODS
Down-conversion sources. The 532-nm pump beam is generated by a frequencydoubled diode-pumped solid-state laser (Coherent Sapphire SF) and is focused
onto the two periodically poled potassium titanyl phosphate (ppKTP) crystals with
dimension 1 mm 3 2 mm 3 2 mm and poling period 9.675 mm for type-0 phase
matching. The crystals are spatially oriented so down-conversion occurs when the
CW pump beam is horizontally polarized (both the signal and idler produced are
also horizontally polarized). In order to conform to the phase-matching conditions
for 810-nm and 1,550-nm photons, NL1 (NL2) is heated to 83.7 uC (84.7 uC). When
the set-up is adjusted to produce 820-nm and 1,515-nm photons (to be used with the
fused silica phase object), NL1 (NL2) is heated to 39.2 uC (39.7 uC). All images were
obtained with 150-mW pump power.
Wavelength filtering. Inside the interferometer, D1 is used to separate the 810-nm
photons from the 1,550-nm photons. Mirror D1 (and also D2) reflects about 93% of
1,550-nm light and transmits about 99% of 810-nm light. Most of the pump beam
going through NL1 is transmitted through both D1 and D4 (each with a transmittance of around 97% at 532 nm) and therefore almost never reaches BS. The dichroic
mirror D5 additionally transmits some 532-nm light (around 25%), so some of the
pump beam that goes through NL2 as well as some of the remaining pump beam
from NL1 are discarded there. All remaining pump beam light is eliminated with
either filters or the imaging object itself. The silicon sample is opaque to both 532-nm
and 810-nm light, thus completely blocking these wavelengths along the path D1
D2. When the other samples are used, a long-pass filter is placed just before the object
to cut out these lower wavelengths. The remaining 532-nm light that is not separated
out through the dichroic mirrors or object is blocked in front of the camera by three
filters. A 3-nm narrowband filter centred at 810 nm and two long pass filters were
attached directly to the front of the camera. As it utilizes a silicon-based detector, the
camera (Andor Luca-R EMCCD) does not detect 1,550-nm photons. Nonetheless, a
combination of spectral filters guarantees that neither 1,550-nm photons nor 532-nm
pump photons reach the camera.
Imaging lens systems. As it is crucial that the down-converted photons be identical,
we use confocal lens systems to image plane 1 onto plane 3 (see Fig. 2), thus ensuring
that the pump beams at NL1 and NL2 are identical, the 810-nm photons when they
combine at the BS are identical, and the 1,550-nm photons are identical from NL2
onward. Lenses L2 and L29 image plane 1 of the pump onto plane 3, and similarly L3
(L4) and L39 (L49) image plane 1 onto plane 3 for the 1,550-nm (810-nm) photons.
Lenses L5 and L6 in combination with L49 image plane 2 onto the EMCCD camera.
Lenses L2, L29, L3, L39, L4, L49 have a focal length of F1 5 75 mm. The distance from
plane 1 to each of L2, L3 and L4 is 75 mm; from those lenses to plane 2 is another
75 mm; from plane 2 to L29, L39 and L49 is also 75 mm; and from those lenses to plane 3
is yet another 75 mm. This ensures that the photons produced in both crystals have
the same waist and divergence when they reach the BS. Lenses L5 and L6 have a focal
length of F2 5 150 mm. They are placed 150 mm after plane 3 and 150 mm before the
camera. The total imaging magnification from the object to the camera is given by
F2 ls
, where ls and li are the wavelengths of the signal and idler photons, respectively.
F1 li
Optical path lengths. In our single photon interferometer the paths D1D4BS
and D1D2BS need to be equal, even though no detected photons actually follow
the entire path D1D2BS. To assure indistinguishability of the emission in the two
crystals (NL1 and NL2) the time delay between the arrival of the signal and idler
for each of the two crystals must be the same. The path length difference between
the signal and idler for the pair from NL1 is the distance NL1D1D2BS subtracted
from the distance NL1D1D4BS. The path length difference between the signal
and idler for the pair from NL2 is zero since the down-conversion is collinear. Thus,
we see that the optical path lengths between D1D4BS and D1D2BS must be
equal to within the coherence length of the photons. The coherence length of the
photons is in our case determined by the filtering (3 nm), so we approximate the
coherence length to be 0.2 mm. The other relevant optical path lengths are the paths
PBSD1D2NL2 and PBSM1NL2. The differential distance between these paths
must be within the coherence length of the laser, which in our case is approximately
200 m.
Intensity object. Our intensity object is constructed from 0.33-mm-thick card stock
with images defined by laser cutting. The images on the object were each 3 mm high.
Microfabricated silicon phase sample. The first custom phase sample consists of
500-mm-thick double-side polished (100)-oriented single-crystal silicon with imaging
targets defined on one face using standard microfabrication techniques. The absorption coefficient of silicon is ,1,000 cm21 at 810 nm (ref. 22), and it is ,1024 cm21
at 1,550 nm (ref. 23). Processing begins by cleaving a 75-mm-diameter silicon wafer
to obtain chips with lateral dimensions of 25 mm 3 25 mm. The cleaved chips are

patterned using conventional optical contact lithography followed by plasma etching. In order to generate a relative p-phase shift at 1,550 nm, features are etched to
a depth of approximately 310 nm (nominal height of 321 nm using a refractive index
of silicon of 3.48; ref. 24) into the exposed Si surface using a cryogenic (2108 uC) SF6/
O2 reactive-ion etching (RIE) process protected with a positive photoresist mask. To
improve thermal transfer, the silicon chips are mounted on a carrier wafer using a
thin layer of vacuum grease. Additionally, in order to minimize variations in the
overall etch depth and thus resulting phase shift from the imaging targets, the feature
linewidth is kept constant over the lithographic pattern to mitigate the effects of
aspect-ratio dependent etching (or RIE lag). After etching, the chips are removed
from the carrier wafer and the masking resist and mounting film are stripped using
a combination of organic solvents and oxygen plasma ashing. In order to eliminate
spurious reflections from the polished surfaces, a dual-sided silicon nitride antireflection (AR) coating is deposited via plasma-enhanced chemical vapour deposition (PECVD) using He-diluted SiH4 and NH3 as reactive process gases. The deposition process yields quarter-wave optical thickness layers at a target film thickness of
(with a refractive index of 1.9 at the imaging wavelength of 1,550 nm).
2,040 A
In order to achieve the highest contrast, the relative path-length difference between
the etched and non-etched regions should be equal to a half wavelength of 1,550-nm
light adjusted for the difference in the indices of refraction of silicon and air. This gives
a target thickness difference of 321 nm (for a refractive index of silicon of 3.48). Given
the slight error in etch depth, the actual thickness difference is 310 nm, which is still
sufficient to obtain high contrast images.
Microfabricated fused silica phase sample. Similar to the silicon phase object described above, the fused silica phase sample, cleaved from a 500-mm-thick glass wafer,
is constructed via a standard lithographic and reactive ion etching process. In this
case the same mask pattern is once again defined with contact lithography. In order
to transfer the features into the fused silica, a high-power inductively coupled plasma
(ICP) RIE process is required (150 W ICP, 250 W RF powers) with an etch chemistry
consisting of SF6 and Ar. Given the poor selectivity to the masking resist, a thick
(10 mm) coating of AZP4620 photoresist is required. The target etch depth of
1,788 nm is achieved within roughly 10 min at room temperature. Given the high
plasma energy, thermalization with the cooled carrier wafer is key. Due to nonuniformities in thermal contact with the carrier, we observe significant variation in
etch depth (6200 nm) across the surface of the 25 mm 3 25 mm pattern. No AR
coating is employed given the small Fresnel reflection (4%) from the low-index silica
substrate.
For 820-nm light, an exact 2p phase shift is given by a thickness difference of
1,811 nm (using an index of refraction of 1.45)25; after processing, the average etch
depth recorded for the fused silica sample is 1,803 nm.
Interference visibility. In order to quantify the visibility in our imaging experiment,
we detect the total intensity of 810-nm photons at one output of BS as a function of
the relative phase between the pump beams that illuminate each crystal. Extended
Data Fig. 1 shows a plot of the count rate measured with an avalanche photodiode
when no object is present. The red circles show the experimental points, and the best
fitting sinusoidal function (red line) gives a visibility of (77 6 1)%. The visibility for
our experiment is given not only by losses in both the 1,550-nm and 810-nm arms of
the interferometer, but also by residual imperfections in the alignment for the two
idler beams. The blue squares correspond to data obtained when the path NL1NL2
is completely blocked, which results in zero interference visibility. Interference only
arises if the idler between the two crystals is unblocked, for only then is its source, and
therefore also the source of its signal sister, unknowable.
Showing that induced emission is negligible in the experiment. In order to
demonstrate in our experiment that the 1,550-nm photons from NL1 do not induce
down-conversion in NL2, we show in Extended Data Fig. 2 the count rates for 810-nm
photons originating at NL2 when the 1,550-nm beam between D1 and D2 was blocked
(blue crosses) and unblocked (red dots). The mean count rate and the standard
deviation were obtained by analysing data obtained over 40 s. The blue diamonds
show that the ratio of the count rates for the blocked and unblocked configuration is
very close to 1 irrespective of the pump power.
22.
23.

24.
25.

Jellison, G. E. Jr. & Modine, F. A. Optical absorption of silicon between 1.6 and 4.7
eV at elevated temperatures. Appl. Phys. Lett. 41, 180 (1982).
Khalaidovski, A., Steinlechner, J. & Schnabel, R. Indication for dominating surface
absorption in crystalline silicon test masses at 1550 nm. Class. Quantum Grav. 30,
165001 (2013).
Malitson, I. H. Interspecimen comparison of the refractive index of fused silica.
J. Opt. Soc. Am. 55, 1250 (1965).
Bass, M. Handbook of Optics Vol. 2, 2nd edn (Optical Society of America, 1995).

2014 Macmillan Publishers Limited. All rights reserved

RESEARCH LETTER

Extended Data Figure 1 | Visibility of the experiment. The count rates were
recorded with the path D1D2 both unblocked (red dots) and blocked (blue
squares) as the relative phase between the transmitted and reflected beams of

the PBS was varied. The red line is a sine curve fit for the experimental data
giving (77 6 1)% visibility. The error bars are smaller than the size of the data
points.

2014 Macmillan Publishers Limited. All rights reserved

LETTER RESEARCH

Extended Data Figure 2 | Excluding induced emission. Shown are the

count rates for 810-nm photons produced in NL2 when the path between D1
and D2 was blocked (blue crosses) and unblocked (red dots). The blue

diamonds show the ratio of the count rates for the blocked and unblocked
configuration. The linear fit for this data (black line) gives an angular coefficient
of (2 6 4) 3 1025 (mW)21.

2014 Macmillan Publishers Limited. All rights reserved